This population-based retrospective cohort study included all live births in Ontario hospitals between April 1, 2007, and March 31, 2018, born to mothers aged 15 to 50 years who were continuously eligible for the Ontario Health Insurance Plan (OHIP) from three months prior to conception until delivery. Continuous OHIP eligibility reflects residency in Ontario and meeting provincial requirements for physical presence during the study period. We used linked health administrative databases at ICES (https://www.ices.on.ca/) to identify opioid exposure during pregnancy and evaluate long-term health outcomes in children, based on maternal-child health service encounters.

The cohort was identified from the Ontario birth registry, the Better Outcomes Registry & Network (BORN). Births occurring between April 1, 2007, and March 31, 2012, were identified from the BORN Niday Perinatal Database, a historical subset of the BORN registry capturing hospital births up to 2012. Births occurring between April 1, 2012, and March 31, 2018, were identified from the BORN Information System (BIS), which is the current perinatal database. For this study, infants were followed up until March 31, 2020 (up to 13 years), or until they lost eligibility for provincial healthcare services, by linking their birth records to healthcare administrative databases [28].

We first excluded stillbirths and maternal and infant records with linkage errors or warnings. Children born to non-Ontario residents, those without a valid health card number, those with missing or implausible gestational age and birth weight, either infants or mothers with an invalid identifier and mismatched birthdates between the Registered Persons Database (RPDB) and BORN and those without healthcare eligibility at birth or within 60 days after birth were excluded. Children with an invalid date of death from the RPDB and those with zero follow-up time were excluded (Figure 1). A timeline of exposure, covariate, and outcome ascertainment is shown (Supplementary Appendix 1). We report this study according to the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) reporting guideline [29].

BORN Ontario (https://www.bornontario.ca/) is the provincial pregnancy, birth, and childhood registry [30]. The Niday and BIS databases in BORN capture maternal-newborn records from hospital births of at least 500g and at least 20 weeks of gestational age in Ontario, with >99% coverage of births in the province. The routine data collection in BORN includes maternal demographics, health behaviours, prenatal screening, pregnancy and obstetric complications, birth outcomes, and newborn outcomes. BORN Ontario applies a comprehensive framework, including both internal and external audits, to maintain quality in all aspects of the database and ensure the reliability, accuracy, validity, and completeness of the data1567424698 [31–34]. We used this database to establish the study cohort and to ascertain prenatal opioid exposure and covariates regarding pregnancy and birth information.

After establishing the cohort, the following databases were linked to identify study outcomes and to supplement exposure and covariate data: the Ontario Health Insurance Program (OHIP) Claims Database, capturing outpatient physician visits through physician billings, the Canadian Institute for Health Information (CIHI) Discharge Abstract Database (DAD) to capture inpatient hospital admissions, the CIHI National Ambulatory Care Reporting System (NACRS) to capture emergency department visits, and the Same-Day Surgery (SDS) database to collect data on outpatient surgical procedures. The Ontario Mental Health Reporting System (OMHRS) was used to acquire information on inpatient admissions to mental health facilities during pregnancy. The Registered Persons Database (RPDB) eligibility data set was used to identify participants with active healthcare coverage. Vital status and length of follow-up for infants were also obtained from the RPDB. The full list of administrative databases that were used in this study, including additional ICES-derived databases, is detailed in Supplementary Appendix 2.

Mother-infant records from the Niday/BIS Perinatal Database were linked to other health services and administrative databases using unique encoded identifiers and analysed at ICES. This independent and non-profit research entity is authorised to collect and store health administrative data for health system evaluation and improvement under Section 45 of the Personal Health Information Protection Act in Ontario. A unique encoded identifier, the ICES Key Number, is assigned to each record based on direct personal identifiers, enabling records to be deterministically linked across datasets while protecting privacy and confidentiality [35].

Prenatal opioid exposure was defined as any indication of maternal opioid use from three months before the estimated date of conception through the date of delivery, capturing opioid exposure across the entire pregnancy period. Prenatal opioid exposure was ascertained through provider-documented data collected during routine prenatal care in BORN or through maternal opioid-related health services records identified in DAD, OHIP, NACRS, or OMHRS. This definition includes any illicit, prescribed, or medication for opioid use disorder (methadone or buprenorphine). BORN collects information on substance or medication use, including opioids, at the first prenatal visit using a standardised perinatal record. A previous re-abstraction study found that the opioid use reported in BORN Ontario had a sensitivity of 84% (95% CI: 60–97%) and a specificity of 98% (95% CI: 91%–99%) compared to clinical records [3].

To identify additional maternal opioid exposure, we used supplemental databases to identify opioid-related health service encounters, including diagnoses of opioid use, abuse, or dependence; opioid poisoning; adverse drug reactions; multiple drug dependence involving opioids; and monthly management under an opioid agonist maintenance programme. These encounters were identified using ICD-10, DSM-IV, and OHIP billing codes from the DAD, OHIP, NACRS, or OMHRS databases (Supplementary Appendix 3). The Narcotics Monitoring System (NMS) database, which provides data on prescribed opioids after July 2012, was used for sensitivity analyses (Supplementary Appendix 8).

The Canadian Paediatric Society recommends routine child examinations in the first week after discharge from birth, at 2, 4, 6, 9, 12, 18, and 24 months, and then once every year until five years of age [36]. The primary outcome was the uptake of well-child visits from birth until 24 months of age. The frequency of well-child visits, including the enhanced 18-month well-child visit, was determined using the diagnostic or billing codes from outpatient physician visits (Supplementary Appendix 3). A well-child visit was defined as any check-up, assessment, or periodic health visit in an outpatient setting. To specifically assess receipt of the enhanced 18-month well-child visit, recommended between 17-24 months and considered a critical period for developmental surveillance, we captured the presence of any preventive visit between 515 and 761 days after birth. For this outcome, the cohort was restricted to children with continuous OHIP eligibility from birth until 761 days (Supplementary Appendix 1).

Secondary outcomes included all-cause use of health services. The rates of all-cause inpatient, outpatient, and emergency department visits were examined to determine the pattern of health services use throughout the entire follow-up period and within specific time intervals: 6 and 12 months of age, 2 years, 3 years, and between 3 years and 13 years of age. The outpatient encounters were examined by comparing primary care and specialty care. Primary care visits were ascertained from the OHIP database. They were defined as any outpatient visits billed by a family physician or a paediatrician with at least one of the OHIP billing codes associated with primary care. Specialist visits were defined as any OHIP encounters billed by specialists, excluding primary care visits billed by a paediatrician. The number of inpatient hospitalisations and emergency room visits was determined from DAD and NACRS encounters, respectively. If an emergency department visit led to hospitalisation, it was counted as one emergency department visit and one hospitalisation to reflect the frequency of service use.

Potential covariates were identified a priori through a literature search. These included maternal age at delivery, neighbourhood marginalisation index, income, adequacy of prenatal care, maternal psychiatric diagnoses, other substance use during pregnancy, maternal pre-existing health conditions, pregnancy complications, congenital anomalies, preterm birth, small for gestational age, admission to the neonatal intensive care unit (NICU) and infant sex. The hypothesised relationships among exposure, outcome, and covariates, along with the minimal sufficient adjustment set, were determined using a directed acyclic graph (Supplementary Appendix 4). Maternal age at delivery was categorised as less than 20 years, 20-29 years, 30-39 years, and 40 years and older. We measured area-based socioeconomic conditions at the census-tract level using the Ontario Marginalization Index [37], a composite index comprised of four dimensions that estimate marginalisation, households and dwellings, material resources, age and labour force, and racialised and newcomer populations, based on Census data (Supplementary Appendix 2). Each dimension was compiled and divided into quintiles, representing low (quintile 1) to high (quintile 5) marginalisation. Income is the median household income at the dissemination area level and is categorised into quintiles. We used the Revised Graduated Prenatal Care Utilisation Index adapted for Ontario to determine prenatal care adequacy, which was categorised into 6 groups (intensive, adequate, intermediate, inadequate, no care, and missing care). Maternal psychiatric diagnoses, including substance-related disorder, mood affective/anxiety disorder, personality and behaviour disorders, and schizophrenia, delusional and psychotic disorders, were each dichotomised (yes or no). Other substance use during the pregnancy was dichotomised by substance and included use of cannabis, tobacco, alcohol, cocaine, hallucinogens, other prescriptions and other substances. Supplementary Appendix 5 provides a comprehensive list of covariates, including the selected confounding variables used in the study, as well as the databases employed to ascertain them.

The proportion of missingness for individual variables included in the analyses varied from <0.1% (n=11, small for gestational age) and 5.5% (tobacco, cocaine, and hallucinogen use during pregnancy), and 10.4% of records had missing information for at least one covariate. We assumed a missing at random (MAR) mechanism for the missing data and used a fully conditional specification approach to impute missing values. We first compared sociodemographic and clinical variables between individuals with complete and incomplete data. Missingness was associated with several observed characteristics, which supports the use of multiple imputation under the MAR framework, and these factors were included in the imputation model. We used the PROC MI procedure in SAS to generate 10 imputed datasets; the list of variables included in the imputation model is provided in Supplementary Appendix 5. In addition, exposure, mediator (preterm birth), and primary outcome (all well-child visits until 24 months of age) variables were included in the imputation model. For the main analyses, statistical models were repeated on each imputed dataset, and parameter estimates and standard errors were combined to produce pooled estimates of association with a corresponding 95% confidence interval (CI).

Baseline characteristics were compared between the two exposure groups using standardised mean differences (SMD). An absolute SMD above 0.1 was considered an imbalance and a meaningful difference of covariate distribution between the exposure groups [38]. Poisson regression models (or negative binomial models when overdispersion was present) were used to estimate the incidence rate of well-child visits up to 24 months and all-cause health services use, with the natural logarithm of follow-up time for each outcome serving as an offset variable. The incidence rate ratio and corresponding 95% CI were calculated and reported. We used a log-binomial model to estimate risk ratios and 95% CIs for the receipt of an 18-month well-child visit between 17 and 24 months of age. All multivariable models were adjusted for the minimal sufficient adjustment set identified from a directed acyclic graph (DAG), including maternal age, neighbourhood marginalisation (ON-Marg), neighbourhood income, adequacy of prenatal care, maternal psychiatric diagnoses, and substance use during pregnancy.

Selected main analyses (uptake of well-child visit during the 18-month interval, incidence rates of well-child visits and specialist visits until the age of 2 years) were repeated among subgroups to account for preterm birth and multiple gestations. Newborns with prenatal opioid exposure are often born preterm [39], which increases the risk of health consequences in newborns [40, 41]. We restricted the cohort to infants born at ≥37 weeks of gestational age to minimise confounding by preterm birth. To account for multiple births, the cohort was restricted to singletons to control for potential confounding.

Several sensitivity analyses were done to assess the robustness of the results. We used coarsened exact matching (CEM) to match opioid-exposed children to unexposed children across covariates to reduce the imbalance in baseline characteristics between exposure groups. CEM was conducted across 10 multiply imputed datasets to account for missing covariate data, with an average of 10,515 exposed and 693,945 unexposed individuals across these datasets. Further details of the CEM methods are given in Supplementary Appendix 6.

Children exposed to opioids in utero had lower incidence rates of well-child visits until the age of 24 months (incidence rate ratio [IRR]: 0.73, 95% CI: 0.72, 0.73), and the association was attenuated slightly but remained statistically significant after adjusting for confounders (adjusted incidence rate ratio [aIRR]: 0.82, 95% CI: 0.81, 0.83) (Table 2). Between 17 to 24 months of age, 67.8% of exposed children received any well-child visits, and they were 0.89 times as likely to receive any well-child visit between 17 to 24 months (adjusted risk ratio [aRR]: 0.89, 95% CI: 0.88, 0.90), adjusted for confounders (Table 3). When the definition of well-child visit was narrowed to the enhanced well-child visit, and using a new fee code implemented in May 2008, the uptake lowered among both exposed and unexposed children (36.1% vs 55.9%), and the risk ratio was also lowered to 0.82 (95% CI: 0.80, 0.84). We used CEM to match and weight the cohort across covariates and to reduce residual confounding. After these adjustments, the association between opioid use and preventive care visits was attenuated towards the null, although it remained statistically significant (Supplementary Appendix 6).

Rates of well-child visits and primary care visits were significantly lower among exposed children than unexposed children throughout the follow-up period (aIRR: 0.92, 95% CI: 0.91, 0.93; Figure 2). The associations remained unchanged until 0-3 years (aIRR: 0.84, 95% CI: 0.83, 0.85) and attenuated between 3 to 13 years (aIRR: 0.90, 95% CI: 0.88, 0.92). Exposed children had a higher incidence rate of specialist visits at the intervals between 0 to 3 years (aIRR: 1.54, 95% CI: 1.48, 1.59), and the associations attenuated progressively with age. Between 3 to 13 years, the specialist visits and prenatal opioid exposure were no longer associated with statistical significance (aIRR 1.05, 95% CI: 1.00, 1.10).

As seen in Figure 2, the exposed children had higher incidence rates of emergency department visits consistently throughout the follow-up period, regardless of the age. Prenatal opioid exposure was associated with increased rates of hospitalisations throughout the follow-up period, but the associations gradually increased with age. Similarly, rate of same-day surgery increased with age and became statistically significant. The mean length of stay for hospitalisations were between 11.7-13 days for exposed children and 4.5-5.2 days for unexposed children, which persisted until 0 to 3 years of age (Supplementary Appendix 7).

Associations observed in the subgroup analyses of singleton and full-term births were consistent with the main analyses for preventive care outcomes, suggesting that multiple births or prematurity did not confound the associations (Table 4). Compared to the main analysis, the associations between the number of specialist visits and prenatal opioid exposure remained unchanged in singleton births from birth to 2 years of age. When the cohort was limited to infants born full-term, the association with specialist visits increased slightly between birth and 6 months (aIRR 2.14, 95% CI: 2.07, 2.21) and between 6 months and 12 months (aIRR 1.90, 95% CI: 1.84, 1.97). The association became comparable between birth and two years of age (aIRR 1.64, 95% CI: 1.59, 1.70).

In sensitivity analyses comparing alternative exposure definitions, we found no meaningful differences in well-child visit uptake between exposure groups, suggesting robust associations across exposure-ascertainment approaches. For example, the rates of preventive and well-child visits were lower among children with opioid exposure identified in BORN, (aIRR 0.87, 95% CI: 0.86-0.88) and this was similar to those with opioid exposure identified using the original definition + ORT (aIRR 0.86, 95% CI: 0.85-0.87) and the original definition + ORT + NAS (aIRR 0.87, 95% CI: 0.86-0.88).

This study has several limitations. First, outpatient billing codes were used to ascertain well-child visits and immunisation outcomes, which may have led to an underestimation of preventive care adherence among children who use non-traditional healthcare settings. Families affected by prenatal opioid exposure are less likely to have a primary care provider and may rely on community-based services, which may not directly bill, introducing a potential undercount in billing codes. Further, as discussed above, a proportion of infants exposed to opioids may receive care from non-billing primary care settings, which could potentially exaggerate observed differences in well-child and primary care visit rates. Emergency department, hospitalisation, and acute care outcomes would be less susceptible to incomplete capture. Second, area level socioeconomic data were used in place of individual-level data due to limitations in data quality and completeness. Consequently, important maternal predictors such as health literacy, employment status, social support, and marital status could not be directly assessed. Additionally, the study cohort spanned births from 2007 to 2018, while socioeconomic variables were derived from the 2016 Canadian Census, introducing a temporal gap that may have resulted in misclassification. Third, the cohort may comprise heterogeneous subpopulations affected by opioid exposure in varying ways over time and type of opioid exposure. Although we used multiple data sources to broadly capture varying types of prenatal opioid use, we were not able to fully distinguish medication for opioid use disorder and unregulated use across all individuals. Opioid use during pregnancy is partly ascertained using self-reported data recorded at the first prenatal visit, which likely underestimates true prevalence due to stigma and underreporting. In addition, the available data cannot reliably distinguish the type of opioid, dose, timing, or patterns of use across pregnancy. Given evidence that child health outcomes vary by type and context of prenatal opioid exposure, this exposure heterogeneity may have contributed to non-differential misclassification and attenuation of observed associations. Differences in health care engagement and social context related to opioid type may also contribute to additional heterogeneity in the observed preventive and acute care use patterns. Policy changes, such as the 2011 implementation of Ontario’s Narcotics Safety and Awareness Act, may have influenced opioid prescribing patterns and prenatal opioid exposure prevalence. As reported by Corsi et al. [3] and Brogly et al. [1], prenatal opioid exposure decreased since 2012 due to changes in opioid prescribing guidelines to reduce prescribing opioids, analgesics, and cough medicines [2]. However, despite the decreasing trend, the prevalence of presumed illicit opioid use, maternal opioid-related hospital care, and neonatal abstinence syndrome remained largely unchanged [2, 3]. We also attempted to address the impacts of potential temporal changes in the opioid crisis by performing subgroup analyses of children born after July 2012. In this subgroup, the prevalence of prenatal opioid exposure was 1.3% between 2012-2018, compared to 0.99% across the full 2007-2018 period. Finally, our findings may be generalisable only to populations with universal healthcare systems. Insurance coverage has been identified as a key determinant of healthcare utilisation, and US-based studies have indicated that Medicaid eligibility significantly influences service use patterns [45, 46, 60–62]. However, given that over 75% of US children with prenatal opioid exposure are Medicaid-eligible, our findings remain relevant for a substantial proportion of this population [60].

Our findings highlight the need to improve preventive healthcare use among children with prenatal opioid exposure. Future research should focus on identifying and implementing targeted interventions to increase adherence to well-child visits, such as expanding access to home visits, mobile clinics, and group well-child visits. Additionally, addressing healthcare provider biases and enhancing parental health literacy may help mitigate barriers to care. These strategies are crucial for ensuring that children affected by prenatal opioid exposure receive timely preventive care and appropriate healthcare resources.